international journal of intelligent control and …

Performance Analysis Of Frame-Burst-based Medium Access ControlProtocols Under Imperfect Wireless Channels

Yu ZHENG, Kejie LU, Dapeng WU, and Yuguang FANG

Abstract— How to provide high throughput performance is a majorchallenge in the design of MAC protocol for next generation wirelesslocal area networks (WLANs) and wireless personal area networks(WPANs). To achieve this goal, frame-burst-based CSMA/CA protocolshave been proposed recently in both academia and industry. The mainidea of the frame-burst-based protocol is to aggregate multiple upper-layer packets into a larger burst frame at the MAC layer; thus the MAClayer overhead can be substantially reduced. In our previous studies, wehave demonstrated analytically that the frame-burst-based protocol cansignificantly improve the throughput performance of CSMA/CA protocolin the error-free channel condition. Since wireless channels are generallyerror-prone in practice, it is very important to study the performanceof MAC protocols under such condition. In this paper, we addressthis issue and develop an analytical model to evaluate the unsaturatedthroughput performance of the frame-burst-based CSMA/CA protocol inimperfect wireless channel. Numerical results show that, our analyticalmodel is highly accurate. More importantly, our results reveal that,given the traffic load and the bit error rate (BER) of channel, the bestthroughput performance can be achieved through appropriate setting ofburst aggregation policy.

Index Term— WLAN, high data rate, MAC, CSMA/CA, analyticalmodel, throughput.

1. Introduction

Wireless local area networks (WLANs) and wireless per-sonal area networks (WPANs) are commonly regarded as thekey components for the future Ubiquitous Computing Age[6]. In the past few years, WLANs and WPANs, especiallyIEEE 802.11 based WLANs [1], have been widely deployedand have attracted significant attention from both academiaand industry. With the advances in wireless communicationtechnologies such as multi-input multi-output (MIMO) [8] andultra-wide Band (UWB) [13], the next generation WLANs andWPANs are capable of providing high data rate ( �� Mb/s)[9], [14] from the physical layer perspective. Consequently,how to efficiently utilize the high data rate provided byphysical layer becomes a major challenge in the design ofmedium access control (MAC) protocols.

In both WLANs and WPANs, one of the most importantMAC protocols is carrier sense multiple access with collisionavoidance (CSMA/CA). Although CSMA/CA generally workswell in a low data rate (less than 10 Mb/s) scenario1, existingstudies have shown that, in high data rate networks, theefficiency of CSMA/CA protocol will be significantly limitedby various overheads, such as preamble, control messages,packet collision, backoff, and inter-frame-spacing. Moreover,

This work was supported in part by National Science Foundation FacultyEarly Career Development Award under grant ANI-0093241 and the Officeof Naval Research Young Investigator Award under grant N000140210464.

The authors are with the Department of Electrical and Computer Engineer-ing, University of Florida, Gainesville, FL-32611, USA.

Please direct all correspondence to Prof. Yuguang Fang. Tel. (352) 846-3043, Fax (352) 392-0044, Email: [email protected].

1In this paper, we only consider the single-hop scenario, in which any twonodes in the network can directly communicate with each other.

many overheads cannot be avoided in practice. For instance,the duration of preamble for a new protocol must be the sameas that of an old protocol to maintain a backward compatibility.Clearly, in such a case, the overhead problem will becomemore and more serious with the increase of channel data rate.

To reduce the overhead and to improve the throughputperformance of CSMA/CA protocol, a number of schemeshave been proposed recently in both academia [10], [20]–[22]and industry [15], [18]. A common feature of these approachesis to aggregate multiple upper-layer packets in a burst frameat the MAC layer and then transmit these packets in oneframe instead of transmitting them one by one. In our previousstudy [10], we proposed a general framework to provide acomprehensive method for burst transmission control. In thispaper, we will study the performance of a frame-burst-basedMAC protocol within the framework. Particularly, we studythe throughput performance of the protocol under differentincoming traffic load in imperfect wireless channel, which hasnever been addressed in the literature.

The performance of CSMA/CA protocols, particularly IEEE802.11, has been studied extensively in the literature [2]–[5],[7], [12], [16], [17], [19], [23], [24]. To simplify the analysis,most early studies, such as [2], [19], assumed that every nodein the network always has packets ready to transmit, known asthe saturated condition. Although these analysis can lead toinsightful observations, a number of recent studies have beenfocusing on a more general, unsaturated, traffic condition.

In [3]–[5], [7], [24], the authors evaluated the unsaturatedperformance by analyzing the behavior of the MAC proto-col only. Although these approaches have low computationalcomplexity, they generally ignore the impact of MAC layerqueue. The queueing behavior was first studied in [16], [17], inwhich a G/G/1 model is in use. However, due to computationalcomplexity, this model is less accurate since it requires anumber of approximations. Moreover, the model is not ap-plicable when the traffic load is high since the queue size isassumed to be infinite. In [12], [23], the system under studyis decomposed into a queueing subsystem and a service timesubsystem. In addition, iterative algorithms are used in bothof them to evaluate the performance of the whole system. Inboth models, the MAC layer queue is modeled as M/G/1/K.The difference between [12] and [23] is that the former useda simplified Markov-modulated general model to estimate theservice time distribution, while the later derives the servicetime distribution directly through a transfer function approach.

Based on the transfer function approach introduced in[23], we have developed an analytical model to evaluate theunsaturated throughput performance for the frame-burst-basedCSMA/CA protocol in [11], where we assumed an ideal

INTERNATIONAL JOURNAL OF INTELLIGENT CONTROL AND SYSTEMSVOL. 10, NO. 1, MARCH 2005, 43-51

channel condition, i.e., without any transmission error. In thispaper, we further develop the analytical model to take intoaccount the impact of imperfect wireless channel, which ismore practical. Our simulation and numerical results showthat, the proposed analytical model is rather accurate undervarious traffic and network conditions. Moreover, the newanalytical model reveals the relationship between the channelbit error rate and the throughput performance of the frame-burst-based CSMA/CA protocol. Particularly, we show thatan optimum burst assembly policy may exist, which can leadto the best throughput performance under a certain channel biterror rate and traffic load condition.

The rest of the paper is organized as follows. In Section 2,we describe our MAC protocol for high date rate ad hoc net-works. In Section 3, we analyze the unsaturated performanceof the MAC protocol. Simulation and numerical results will beprovided in Section 4. Finally, Section 5 concludes the paper.

2. A Frame-burst-based CSMA/CA Scheme

In this section, we describe the frame-burst-based MACscheme proposed in [10]. The main idea of this scheme isthat a node can aggregate multiple upper-layer packets intoa frame burst at the MAC layer and then transmit the frameburst according to CSMA/CA protocol. Compared to the tradi-tional approach in which upper-layer packets will be deliveredindividually, the proposed scheme can significantly reduce theoverhead, thus can improve the throughput performance ofhigh data rate wireless networks dramatically.

It is important to note that a number of burst aggregationschemes have been proposed in both academia [20]–[22] andindustry [15], [18]. However, the focus of these schemesis the detail of protocols, for example, the frame structureand handshake procedures. In contrast, [10] provides a morecomprehensive framework to support frame burst transmission.Particularly, our framework includes five components:� Packet classification policy specifies the packet classifi-

cation method according to the quality-of-service (QoS)requirement and destination of incoming packets. Notethat packets will be queued in buffers after classification.� Buffer management policy manages the buffer in a waythat achieves QoS requirements and/or fairness amongdifferent flows.� Packet assembly policy determines when and how toassemble a frame burst. This policy should take intoaccount synchronization overhead, physical layer con-straints, QoS, and fairness among different nodes.� Acknowledgment policy defines the procedure of ac-knowledgment at the receiver side.� Packet error control policy dictates what error controlscheme to be used, since packet errors are unavoidable.

In the rest of this section, we first review the classicCSMA/CA protocol. We then provide a simple burst assemblypolicy that will be used in our analysis. Finally, we addressthe burst retransmission policy issue, which will be affectedby the packet transmission errors.

2.1. A Classical CSMA/CA Protocol

In this paper, we consider a well-known CSMA/CA pro-tocol, IEEE 802.11 distributed coordination function (DCF)[1]. To simplify the discussion, we assume that all packetsin a frame burst have the same destination. Therefore, mostexisting functions of IEEE 802.11 DCF can be re-used. Belowwe briefly describe the binary exponential backoff scheme inDCF. Details of other functions in DCF can refer to [1].

Generally, in IEEE 802.11 DCF, a node with packet to sendwill randomly choose a backoff counter if the channel is busy.The backoff counter is uniformly chosen in the range of [0,CW-1], where CW is called contention window. The size ofcontention window is defined as follows. Initially, CW is setto the minimum value, denoted as CWmin. After every un-successful transmission, if the total number of retransmissionis less than a predefined retry limit

�, then the value of

CW will be doubled until it reaches the maximum contentionwindow size CWmax (=CWmin �� ). Now let � =CWmin,then the contention window size of the -th transmission canbe defined as

�� (1)

The counter will be decreased by one if the channel issensed idle for a certain time unit � , the value of whichdepends on the physical layer specification. Retransmissionwill take place when the counter reaches zero. If a data frameis successfully received by the destination node, or if it is notsuccessfully received after

�times of retransmission, the data

frame will be discarded at the MAC layer of the source node.In this paper, we define departure as the event of a data framebeing removed from the MAC layer of the source node for theabove two reasons.

2.2. Burst Assembly Policy

In the analysis, we consider a simple burst assembly policyas follows. In this scheme, we assume that there is only oneclass of traffic from upper layer. Therefore, packets from upperlayer will be queued according to their destinations. Specifi-cally, we need � packet queues in each node, where � is thenumber of nodes in the network. Among the queues, �� are used for buffering packets destined to other �� nodesrespectively, and one queue is used for buffering broadcastpackets. For each queue, we use tail-dropping when there is abuffer overflow.

A frame burst will be assembled if the total number ofpackets in a packet queue exceeds a threshold � �� and thetransmission buffer is empty. This transmission buffer is usedto store the single burst in service and is shared by all thequeues. In addition, we assume that the total number of packetsin a burst must be no more than a preset value �� "! .

2.3. Packet Error Control Policy

In wireless networks, packet transmission errors due tovaried channel conditions are unavoidable and unpredictable.Therefore, error control scheme is required to mitigate the

Zheng et al.: PerformanceAnalysis Of Frame-Burst-based Medium Access Control Protocols... 44

impact. In IEEE 802.11 DCF, this task is achieved by the pro-cedure of packet retransmission, which is basically a stop-and-wait Automatic Repeat Request (ARQ) protocol. In the frame-burst-based MAC protocol, the source node may retransmitonly the packets that were not received correctly in one frameburst. However, such a scheme requires the modification ofMAC protocol to identify the error status of a certain packetwithin a frame burst. It also requires the receiver to maintainreceived packets in buffer to keep the packet order. In short,from a practical point of view, there is a trade-off betweenefficiency and cost in the design of packet error control policy.In this paper, as the first step of our research, we consider asimple policy, in which a frame burst will be retransmitted ifthere is any bit error in the transmission.

3. Analytical Model

In this section, we develop an analytical model to evaluatethe unsaturated throughput performance of the frame-burst-based MAC protocol in imperfect wireless channel conditions.Following [11], we first decompose the whole system into aqueueing subsystem and a service time subsystem.

Since a burst will be retransmitted if the previous transmis-sion attempt failed, the queueing model remains the same asthat of [11], which is an M/G � �� /1/K queue. In thisformulation, � is the capacity of the queue excluding packetsin the burst buffer, and the superscript � � �� !�� meansthat the total number of packets in a burst is an integer in therange of � � �� "!�� .

The main difficulty of our analysis is how to take intoconsideration the impact of transmission errors, which will di-rectly affect the formulation of the binary exponential backoffscheme and the burst service time distribution.

The rest of this section is organized as follows. In subsec-tion 3.1 and 3.2, we provide the assumptions and notationsused in our analysis. We then elaborate on the impact oftransmission error in modeling of binary exponential backoffand the service time distribution, in subsection 3.3 and 3.4,respectively. The throughput analysis will be given in subsec-tion 3.5. And finally, we describe the iterative algorithm forthe analysis in subsection 3.6.

3.1. Assumptions

To conduct the analysis, we first make the following as-sumptions:� There are � identical nodes in the network.� At each node, packet arrivals are Poisson with rate �

(packets/sec), and all incoming packets to a node havethe same destination. Therefore, the burst assembly policycan be simplified as follows:Suppose there are � packets in the queue just before aframe burst departure, then the number of packet in thenext frame burst, denoted as �� , is defined as

� � �� !� ! � ! � � � � � (2)

� The size (in bytes) of incoming packets is an i.i.d randomvariable with an arbitrary distribution �! #"�$ . We furtherdefine � �� and � � "! be the minimum and maximumsize of a packet in bytes.� The queue of MAC layer has the capacity � (in packets),excluding the transmission buffer.� The channel is not perfect in that every bit in a frameburst encounters error with a probability % . To simplifythe discussion we also assume that control packets andframe headers are error free.� The burst service time is an integer multiple of a presettime unit & (in seconds). This integer has an upper bound' � "! as a server only tries to send one burst for finitenumber of times and each time the attempt has a finiteduration.� The probability that a burst transmission attempt fails,denoted as ( , does not depend on the backoff stage ofthe node.� The propagation delay is negligible.

3.2. The Notations

In this section, we provide the notations for the parameterswe will use in the following sections:� (*) denotes the probability that a node is idle at the

beginning of a time segment2, i.e. the node has no burstto transmit in the time segment.� (�+ ( � �� -, � � ! ) denotes the probability that aframe burst consists of , packets.� (/. denotes the probability that a node will transmit in onetime segment.� (�0� denotes the steady state probability that there are �packets left in the queue at the time instance just beforea burst departures.�21 + � denotes the steady state probability that the burstservice time is 34& , given that there are , packets in theburst.�65 denotes the MAC throughput per node in bits persecond.

3.3. Binary Exponential Backoff

To analyze the exponential backoff scheme of theCSMA/CA protocol, we will use the Markov modeling tech-nique introduced in [2], [19]. Particularly, we can first partitionthe continuous time axis into time segments, where twoconsecutive segments are delimited by the event of a valuechange in the backoff counter. We can then formulate a two-dimensional discrete time embedded Markov chain as that of[2], [19]. With the close form solution of the Markov chain,the first relation between ( and (/. can be derived by Eq. (3).

The second relation comes from the physical meaning of( . Since a successful burst delivery only happens when thereis neither collision nor bit error in a transmission attempt, wecan calculate ( through( � � �� (*78$9 � �:(�;9$<� (4)

2In this paper, time segment is defined as the time duration in which thebackoff counter of a node does not change. More detail explanation can befound in Section 3.3.

45 INTERNATIONAL JOURNAL OF INTELLIGENT CONTROL AND SYSTEMS, VOL. 10, NO. 1, MARCH 2005

( . �� 9(*$9 � �:( � �� $ � � �<(*$ � ��( � �� $�� (*$9 � �6 �<(�$ � �� $ � � ��

� � � �9(*$ � ��( � �� $ � � �<(*$ � ��( � �� $�� (*$9 � �6 �<(�$ � � �� $�� ( � � �� <(�$9 � ��( �� $ � � �� (3)

where ( 7 is the collision probability and ( ; is the average bursterror probability. (/. and ( ; can be expressed as

(�7�� 6� � �6 � �:(*) $#( . � � � (5)

and

( ; � �� + � �� (*+�� ( ; , $<� (6)

where we let ( ; , $ be the probability that the burst transmis-sion is not successful due to bit error, given that there are ,packets in the burst.

To calculate (�; , $ , we first define �! , � "�$ as the probabilitydistribution function of the size (in bytes) of a burst with, packets. Since the length of incoming packets is i.i.d.random variable, we can achieve �! , � "�$ through an recursivealgorithm

�! � � "�$ � �! "�$�! � � "�$ � �! "�$�� ! � � "�$��! , � "�$ � �! "�$�� ! , � � � "�$<� (7)

where � represents the convolution. Consequently, ( ; , $ canbe calculated by

( ; , $�� +�� +�� ! , � "�$ � � �� $�� (8)

3.4. Service Time Distribution

To derive the service time distribution, we use the techniqueintroduced in [11]. First, we define � + �� $ as the probability-generating function (PGF) of 1 + � , which is� + �� $ � �

� � � � 1 + � � (9)

Clearly, � + �� $ is the z-transform of 1 + � and there is a one-to-one correspondence between � 1 + �� and � + �� $ . Next, welet � � be the duration of segment " and let � �� be theduration within segment " , when the server is busy. We canthen apply the transfer function approach, in which the MAClayer transmission process is characterized by a linear system,as shown in Fig. 1.

The parameters used in Fig. 1 are defined as the following� ( , $ denotes the average probability that the transmissionof a given burst with , packets is not successful in onetime segment. Similar to Eq. (4), we have

( , $ � � �6 � �:( 7 $ � ��( ; , $ $ (10)

�! "�$ denotes the PGF of � �� given that the currentnode has a packet in the transmission buffer but nottransmitting.

�!# + �� $ denotes the PGF of � �� given that a collisionoccurs when the current node is transmitting a burst with, packets.�65 + "�$ denotes the PGF of � �� given that the current nodehas successfully transmitted a burst with , packets.�$ �� $ denotes an interim function �� $ � �

��&% � � �� $�� (' "�$��)��*� ,+ � � � �� $.-!�(11)

which is used to simplify the notation.

The formulation of � + "�$ can be directly derived fromFig. 1, as

� + �� $ � � ��( , $ $ 5 + "�$ �� 0/ 1 2 + "�$ $ 3� �0/ � �� $ 4� �2*+ �� $ $ � �� 3� �0/ � �� $<�

(12)where 2 + "�$ � (*7 # + "�$�� ( , $ �:(�7 $ 5 + �� $ � (13)

In the remaining of this subsection, we will discuss how tocalculate 5 + �� $ , # + "�$ , and "�$ numerically.

To calculate 5 + "�$ , we can use

5 + �� $ � +�� +�� ! , � "�$5��6�78 �:9<;>="? � � �A@CB�D�EGF � (14)

where H 0 is the physical layer data rate and I�J5K is the timeoverhead for a successful frame transmission. The calculationof I J�K depends on the detailed protocol specification. Forinstance, I J5K for the RTS/CTS access scheme can be found in[11].

To simplify the discussion, we provide the calculation of# + "�$ and "�$ only for the RTS/CTS scheme. For theRTS/CTS scheme, since collision can only occurs when twoor more RTS packets collide, # + �� $ can be derived as# + "�$ �L��6NM�O ?8 F (15)

where I�7 K is the time overhead for collision.To calculate �� $ , we can use �� $ � � � 1 . $5� 6NP8 F � 1 J 5 �� $:� 1 . � 1 J $ # �� $<� (16)

where 1 . denotes the probability that there is at least onetransmission burst in � � � neighbor nodes in segment " , 1 Jdenotes the probability that there is only one burst transmissionin � � � neighbor nodes in slot " , 5 "�$ denotes the PGF of � ��given that there is a successful transmission in time segment " , Yu Zheng et al.: PerformanceAnalysis Of Frame-Burst-based Medium Access Control Protocols... Yu Zheng et al.: PerformanceAnalysis Of Frame-Burst-based Medium Access Control Protocols... Zheng et al.: PerformanceAnalysis Of Frame-Burst-based Medium Access Control Protocols...


1−p(b)

0 H (z)1 H (z)M

S (z)b

C (z)bQ (z)bC (z)b

S (z)b

C (z)b

S (z)b

C (z)b

S (z)b

pc pc pc pc

S (z)bp(b)−pc p(b)−pc p(b)−pc p(b)−pc

1

1−p(b) 1−p(b) 1−p(b)

H (z)

(a) � + "�$1

H(z) H(z)

1/Wm

H(z)

1/Wm 1/Wm 1/Wm 1/Wm

1 1 1 1H (z)m

(c) �� $Fig. 1. Service system diagram.

and # �� $ denotes the PGF of � �� given that there is a collisionin segment " . These parameters can be calculated through

1 . � � �� 6 � �:(*) $#( . � � � (17)1 J � � � � $9 � ��( ) $ (/.9� � �� ( ) $ (/. � � ' (18)

5 �� $ � �� + � �� (*+ � 5 + "�$ (19)# �� $ � # + �� $ � (20)

3.5. Throughput Analysis

Now let � � �� $ be the total number of burst departures, justbefore which there are � packets waiting in the queue; let� , $ be the departure rate (in bursts / second) of bursts with, packets in the frame. Based on the burst assembly policy,we have

� , $ � ��.��

� �� + � � �� $

� � � �� + ��.��

�� $�

��

�� + ( 0�I J � �

� � �� 0/ � �� $ ( 0�

� ( +I J � �

� � � �� 0/ � �� $ ( 0�

� (21)

We can then calculate the expected size (in bits) of data thatare successfully delivered in time duration �� , given that thesepackets are all conveyed through bursts with , packets in eachburst. Since each burst will be transmitted up to

� � � times,we have

5 , �� $�� , $�� , � � � % � �6 ( , $ $ � �� - � (22)

where � is the average length of packet (in bits), which canbe calculated through

�� " � �! "�$ � (23)

Finally, we can calculate the throughput (per node) of theframe-burst-based CSMA/CA protocol through

5 � �� + � � �� 5 , �� $� �

� � � � �� + � � �� , $ � , � % � �� ( , $ $ � �� - � (24)

3.6. The Iterative Algorithm

In this section, we provide the iterative algorithm to calcu-late the unsaturated throughput of a node:

Step 1: Initialize (*) and ( + to saturated condition, i.e., let(/) � � , ( �� , and ( + � � for ,�� ! .Step 2: Calculate ( and ( . according to an Markov model

for exponential backoff.Step 3: Calculate service time distribution 1 + � through the

transfer-function approach, using ( and ( . .Step 4: Calculate (�0� by using the M/G � � �� /1/K

queueing model.Step 5: Calculate new (�) and ( + based on (�0� .Step 6: Calculate the throughput 5 . Stop the algorithm if5 converges; otherwise go to Step 2 with the new ( ) and (�+

values.It is important to note that, although the convergence of the

iterative algorithm has not been proved, the algorithm seemsalways converge in all our numerical calculations.

4. Simulation and Numerical Results

In this section, we evaluate the performance of the frame-burst-based MAC protocol under various traffic and channel


0

10

20

30

40

50

60

70

80

90

100

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Thr

ough

put (

Mb/

s)

Load (Erlang)

BER=0(Sim)BER=0(Ana)BER=1e-6(Sim)BER=1e-6(Ana)BER=1e-5(Sim)BER=1e-5(Ana)

0

10

20

30

40

50

60

70

80

90

100

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Thr

ough

put (

Mb/

s)

Load (Erlang)


(a) � � �� "!�� (b) � � �� "! � � � ��

0

10

20

30

40

50

60

70

80

90

100

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Thr

ough

put (

Mb/

s)

Load (Erlang)

[1,10] BER=1e-5(Ana)[10,10] BER=1e-5(Ana)[1,10] BER=1e-6(Ana)[10,10] BER=1e-6(Ana)[1,1] BER=0(Ana)

(c) Comparison

Fig. 2. Throughput performance vs. incoming traffic data rate with different channel BER (Scenario (1), �� ).

condition. Moreover, we compare the simulation results tothe numerical results achieved through the proposed analyticalmethod in Section 3.

The basic setting of our experiments is listed as the follow-ing:� All nodes are located in a 10 m � 10 m area.� The size of each packet follows the same distribution.

Unless otherwise specified, we assume that the packetsize is fixed at 1000 Bytes.� Packet arrivals to each node are modeled by a Poissonprocess with the same rate � (packets/s). Consequently,the incoming traffic data rate is H � � � � � (bits/s).� We assume that the preamble overhead, denoted as I�J�� 7 ,is identical for all messages.� We assume that all messages are transmitted with thesame channel data rate H . We then define the traffic loadto the network as H � H Erlang.� We assume that packet transmission error can only occurin payload, with a fixed bit error ratio.� We assume that the RTS/CTS scheme is used with settinglisted in Table I.

In addition, we conduct the experiments in the following

TABLE I

SETTING OF THE MAC PROTOCOL.

Minimum contention window size 8Maximum contention window size 256

Long retry limit 4Queue size 50

two scenarios:1) IEEE 802.11n

In this case, we assume that � � � �� , SIFS = �� ,DIFS = � �� , and H � � � � � , � .

2) High data rate UWBIn this case, we assume that � � � � , SIFS = �� , DIFS= � � , and H � � � � � , � .

Figure 2 shows the throughput performance versus incom-ing traffic rate under different BER conditions in the firstscenario and we let � � � � . In this experiment, we usethree burst assembly setting: 1) � � �� "! �� ; 2)� � �� "!�� , and 3) � � �� !�� . FromFig. 2 (a) and (b), we can observe that our analytical modelis highly accurate under different traffic and BER conditions.We notice that, compared to the benchmark case where BER


0

10

20

30

40

50

60

70

80

90

100

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Thr

ough

put (

Mb/

s)

Load (Erlang)


0

10

20

30

40

50

60

70

80

90

100

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Thr

ough

put (

Mb/

s)

Load (Erlang)


(a) � � �� "!�� (b) � � �� "! � � � ��

0

10

20

30

40

50

60

70

80

90

100

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Thr

ough

put (

Mb/

s)

Load (Erlang)

[1,10] BER=1e-5(Ana)[10,10] BER=1e-5(Ana)[1,10] BER=1e-6(Ana)[10,10] BER=1e-6(Ana)

(c) Comparison

Fig. 3. Throughput performance vs. incoming traffic data rate with different channel BER (Scenario (2), �� ).

is 0, the overall throughput performance decreases slightly ifBER is �� and if the traffic load is high. These results showthat, the burst assembly policy and the simple retransmissionpolicy can work fine if BER is � � �� . One the other hand,if the BER is �� , we can see that the protocol can stillperform well if the load is very small, which suggests thatmore sophisticated retransmission mechanism must be utilizedto improve the throughput performance in a high bit errorchannel condition.

We compare the performance of the three different assemblypolicies under various channel conditions in Fig. 2 (c). Wecan observe that, setting (2) can achieve the best performancein all traffic conditions. Moreover, the results show that theframe-burst-based MAC protocol can significantly outperformthe original CSMA/CA protocol even in imperfect wirelesschannel. Specifically, we note that the throughput performanceof setting (3) in an error-free channel is much smaller than thatof the other two policies with BER as high as � � � � .

In Fig. 3, we show the performance of the frame-burst-based protocol in the second scenario, where we keep allother setting the same as that of Fig. 2. Similarly, we cansee that, compared to the benchmark case where BER is 0,

the overall throughput performance decreases slightly if BERis � � �� and only if the traffic load is higher than � � � Erlang.Interestingly, we found from Fig. 3 (a) that, if the BER is �� ,the maximum throughput is achieved when the load is about� � � � Erlang, which is larger than the saturated throughput.Such results demonstrate that, our analytical model is muchbetter than the saturated model, which can only be used topredict the throughput performance when the traffic load isclose to � .

From Fig. 3 (c) we can see that the performance of policy(2) is better than policy (1) when the BER is �� . However,policy (1) performs better the policy (2) if the BER is � � � �under medium traffic load (i.e., � � � � � �� Erlang).

In Fig. 4 we demonstrate the impact of � ! on theperformance of the protocol with different network traffic loadand BER= � � � � in the second scenario. We can see that, whenthe traffic load is � Erlang, the best throughput performanceis achieved if � � "! is 3; when the traffic load is � � � Erlang,the best throughput performance is achieved if �� ! is 2; andwhen the traffic load is � � � Erlang, the throughput is nearlythe same if � � "! � � , and is larger than that of � ! � � .From Fig. 4 we can also observe that our analytical model


40

45

50

55

60

65

70

1 2 3 4 5 6 7 8 9 10

Thr

ough

put (

Mb/

s)

Bmax

Load=0.5 Erlang(Sim)Load=0.5 Erlang(Ana)Load=0.6 Erlang(Sim)Load=0.6 Erlang(Ana)Load=1.0 Erlang(Sim)Load=1.0 Erlang(Ana)

Fig. 4. Throughput performance vs. �� (Scenario (2), � � �� ,BER= �� ).

0

10

20

30

40

50

60

70

80

90

100

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Thr

ough

put (

Mb/

s)

Load (Erlang)

[5,10], BER=0(Sim)[5,10], BER=0(Ana)[5,10], BER=1e-6(Sim)[5,10], BER=1e-6(Ana)[5,10], BER=1e-5(Sim)[5,10], BER=1e-5(Ana)[1,1], BER=0(Sim)[1,1], BER=0(Ana)

Fig. 5. Throughput performance vs. incoming traffic data rate with differentchannel BER (Scenario (2), � �� ).

is less accurate (over-estimates the throughput) if the loadis � � � Erlang and when � ! is larger than 5. The reasonof this phenomenon can be found in Fig. 3 (a), where wecan observe that, with a slight increase of traffic load, thethroughput decreases significantly from the maximum value.

Fig. 5 shows the performance of the protocol in the scenariowhen the total number of node in the network is �� . Wecan first observe that, for the traditional approach ( � �� ! � � ), the throughput performance will decrease with theincrease of load if the load is larger than � � � Erlang. Similar toprevious results, we show that the performance of the frame-burst-based protocol is better than that of the traditional oneeven if the BER is � � � � .

Finally, Fig. 6 shows the performance of the protocol whenthe size of incoming packets are not fixed. In particular,we assume that the length of incoming packets is uniformlychosen in � � � � � � � � � � Bytes. Comparing the simulation andanalytical results, we can see that our analytical model ishighly accurate under different traffic and channel conditions.

0

10

20

30

40

50

60

70

80

90

100

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Thr

ough

put (

Mb/

s)

Load (Erlang)

BER=0(Sim)BER=0(Ana)BER=1e-5(Sim)BER=1e-5(Ana)

Fig. 6. Throughput performance vs. incoming traffic data rate with differentchannel BER (Scenario (1), � � � � , � � �� ).

5. Conclusions

In this paper, we developed an accurate analytical modelto evaluate the unsaturated throughput performance of aframe-burst-based CSMA/CA protocol in imperfect wirelesschannel. Simulation and numerical results show that, theproposed analytical model is highly accurate in most cases.More importantly, the results reveal that the best throughputperformance can be achieved through appropriate setting ofthe burst aggregation policy.

REFERENCES

[1] Wireless LAN medium access control (MAC) and physical layer (PHY)specifications, 1999. ISO/IEC 8802-11:1999(E).

[2] G. Bianchi. Performance analysis of the ieee 802.11 distributedcoordination function. IEEE J. Select. Areas Commun., 18(3):535–547,March 2000.

[3] Yunli Chen, Qing-An Zeng, and D. P. Agrawal. Performance analysisand enhancement for IEEE 802.11 MAC protocol. In Proc. IEEE ICT,pages 860–867, March 2003.

[4] Yunli Chen, Qing-An Zeng, and D. P. Agrawal. Performance of MACprotocol in Ad Hoc networks. In Proc. Communications Networks andDistributed System Modeling and Simulation Conference, pages 55–63,January 2003.

[5] Yunli Chen, Qing-An Zeng, and D. P. Agrawal. Performance evaluationfor IEEE 802.11e enhanced distributed coordiated function. John Wiley& Sons, Wireless Communications and Mobile Computing, (4):639–653,2004.

[6] Imrich Chlamtac, Marco Conti, and Jennifer J.-N. Liu. Mobile ad hocnetworking: imperatives and challenges. Elsevier Ad Hoc Networks,1(1):13–64, January 2003.

[7] Mustafa Ergen and Pravin Varajya. Throughput analysis and admissioncontrol for IEEE 802.11a. to appear in ACM MONET, 2004.

[8] D. Gesbert, M. Shafi, Da shan Shiu, P. J. Smith, and A. Naguib. Fromtheory to practice: an overview of MIMO space-time coded wirelesssystems. IEEE J. Select. Areas Commun., 21(3):281–302, April 2003.

[9] IEEE P802.15-04/0137r1. DS-UWB physical layer submission to 802.15Task Group 3a. Project: IEEE P802.15 Working Group for WirelessPersonal Area Networks (WPANs), March 2004.

[10] K. Lu, D. Wu, and Y. Fang. A novel framework for medium accesscontrol in ultra-wideband ad hoc networks. Dynamics of Continuous,Discrete and Impulsive Systems (Series B) Special Issue on Ultra-Wideband (UWB) Wireless Communications for Short Range Communi-cations, (to appear), 2005.

[11] Kejie Lu, Dapeng Wu, Yuguang Fang, and Robert C. Qiu. Performanceanalysis of a burst-frame-based MAC protocol for ultra-wideband adhoc networks. In Proc. IEEE ICC 2005 (to appear), May 2005.


[12] M. Ozdemir and A. B. McDonald. A queuing theoretic model forIEEE 802.11 DCF using RTS/CTS. In IEEE Workshop on Local andMetropolitan Area Networks (LANMAN 2004), pages 33–38, April 2004.

[13] S. Roy, J. R. Foerster, V. S. Somayazulu, and D. G. Leeper. Ultraw-ideband radio design: The promise of high-speed, short-range wirelessconnectivity. Proc. IEEE, 92(2):295–311, February 2004.

[14] Status of Project IEEE 802.11n, Online availablehttp://grouper.ieee.org/groups/802/11/Reports/tgn update.htm.

[15] TGn Sync Proposal for IEEE 802.11n, Online http://www.tgnsync.org.[16] Omesh Tickoo and Biplab Sikdar. Queueing analysis and delay mitiga-

tion in IEEE 802.11. In Proc. IEEE INFOCOM, Hong Kong, P. R. China,April 2004.

[17] Omesh Tickoo and Biplab Sikdar. A queueing model for finite loadIEEE 802.11 random access MAC. In Proc. IEEE ICC, pages 175–179,Paris, France, 2004.

[18] World-Wide Spectrum Efficiency, Online http://www.wwise.org.[19] Haitao Wu, Yong Peng, Keping Long, Shiduan Cheng, and Jian Ma.

Performance of reliable transport protocol over IEEE 802.11 wirelessLAN: analysis and enhancement. In Proc. IEEE INFOCOM, pages 599–607, June 2002.

[20] Yang Xiao. MAC performance analysis and enhancement over 100 Mbpsdata rates for IEEE 802.11. In Proc. IEEE VTC Fall, pages 1869–1873,October 2003.

[21] Yang Xiao. Concatenation and piggyback mechanisms for the IEEE802.11 MAC. In Proc. IEEE WCNC, pages 1642–1647, March 2004.

[22] Yuan Yuan, Daqing Gu, William Arbaugh, and Jinyun Zhang. High-performance MAC for high-capacity wireless LANs. In Proc. IEEEICCCN, 2004.

[23] Hongqiang Zhai, Younggoo Kwon, and Yuguang Fang. Performanceanalysis of IEEE 802.11 MAC protocol in wireless LAN. Wiley Journalof Wireless Communications and Mobile Computing (WCMC), pages917–931, December 2004.

[24] Eustathia Ziouva and Theodore Antonakopoulos. The IEEE 802.11distributed coordination function in small-scale Ad-Hoc wireless LAN.International Journal of Wireless Information Networks, 10(1), January2003.

Yu Zheng received her B.S. degree in Electricaland Computer Engineering in July, 1999 from SouthChina University of Technology, Guangzhou, China.She is a graduate research assistant at the De-partment of Electrical and Computer Engineering,University of Florida, and will receive her M.S.degree in August 2005.

Dr. Kejie Lu received the B.S. and M.S. degrees inTelecommunications Engineering from Beijing Uni-versity of Posts and Telecommunications, Beijing,China, in 1994 and 1997, respectively. He receivedthe Ph.D. degree in Electrical Engineering from theUniversity of Texas at Dallas in 2003. He is nowa postdoctoral research associate at the departmentof Electrical and Computer Engineering, Universityof Florida. His research interests include architectureand protocols design for WDM optical networks andwireless networks, performance analysis, network

security, and wireless communications systems.

Dr. Dapeng Wu is currently an Assistant Pro-fessor of Department of Electrical and ComputerEngineering, University of Florida. His researchinterests are in the areas of multimedia, wirelesscommunications, networking, signal processing, andinformation and network security. He received theIEEE Circuits and Systems for Video Technology(CSVT) Transactions Best Paper Award for Year2001. Currently, he is an Associate Editor for theIEEE Transactions on Vehicular Technology andAssociate Editor for International Journal of Ad Hoc

and Ubiquitous Computing. He served as Program Chair for IEEE/ACM FirstInternational Workshop on Broadband Wireless Services and Applications(BroadWISE 2004); and as TPC member of 16 conferences. He is ViceChair of Mobile and wireless multimedia Interest Group (MobIG), TechnicalCommittee on Multimedia Communications, IEEE Communications Society.

Dr. Yuguang Fang received the BS/MS degrees inMathematics from Qufu Normal University, Qufu,Shandong, China, in 1987, a Ph.D degree in Sys-tems and Control Engineering from Department ofSystems, Control and Industrial Engineering at CaseWestern Reserve University, Cleveland, Ohio, inJanuary 1994, and a Ph.D degree in Electrical Engi-neering from Department of Electrical and ComputerEngineering at Boston University, Massachusetts, inMay 1997.

From September 1989 to December 1993, he was a teaching/researchassistant in Department of Systems, Control and Industrial Engineering at CaseWestern Reserve University, where he held a research associate position fromJanuary 1994 to May 1994. He held a post-doctoral position in Departmentof Electrical and Computer Engineering at Boston University from June 1994to August 1995. From September 1995 to May 1997, he was a researchassistant in Department of Electrical and Computer Engineering at BostonUniversity. From June 1997 to July 1998, he was a Visiting Assistant Professorin Department of Electrical Engineering at the University of Texas at Dallas.From July 1998 to May 2000, he was an Assistant Professor in the Departmentof Electrical and Computer Engineering at New Jersey Institute of Technology,Newark, New Jersey. In May 2000, he joined the Department of Electrical andComputer Engineering at University of Florida, Gainesville, Florida, where hegot early promotion to Associate Professor with tenure in August 2003 andhe is Professor since August 2005. His research interests span many areasincluding wireless networks, mobile computing, mobile communications,mobile ad hoc networks, wireless sensor networks, wireless security, automaticcontrol, and neural networks. He has published over one hundred and fifty(150) papers in refereed professional journals and conferences. He receivedthe National Science Foundation Faculty Early Career Award in 2001 andthe Office of Naval Research Young Investigator Award in 2002. He alsoreceived the 2001 CAST Academic Award. He is listed in Marquis Who’sWho in Science and Engineering, Who’s Who in America and Who’s Whoin World.

Dr. Fang has actively engaged in many professional activities. He is asenior member of the IEEE and a member of the ACM. He is an Editorfor IEEE Transactions on Communications, an Editor for IEEE Transactionson Wireless Communications, an Editor for IEEE Transactions on MobileComputing, an Editor for ACM Wireless Networks, and an Editor for IEEEWireless Communications. He was an Editor for IEEE Journal on Selected Ar-eas in Communications: Wireless Communications Series, an Area Editor forACM Mobile Computing and Communications Review, an Editor for WileyInternational Journal on Wireless Communications and Mobile Computing,and Feature Editor for Scanning the Literature in IEEE Wireless Commu-nications (formerly IEEE Personal Communications). He has also activelyinvolved with many professional conferences such as Steering Committee Co-Chair for International Conference on Quality of Service in HeterogeneousWired/Wireless Networks (QShine), QShine’05 (General Chair), QShine’04(Vice General Chair), ACM MobiCom’02 (Committee Co-Chair for StudentTravel Award), MobiCom’01, IEEE INFOCOM’06, INFOCOM’05 (Vice-Chair for Technical Program Committee), INFOCOM’04, INFOCOM’03,INFOCOM’00, INFOCOM’98, IEEE WCNC’00 (Technical Program Vice-Chair), and IEEE Globecom’04 (Symposium Co-Chair).


international journal of intelligent control and …

Documents